How to Enable Meta's Business AI Agent on Your WhatsApp Business API Numbers

The previous post on this blog covered what Meta's Business AI Agent is in plain terms. This one is for the more technical case: if you're already running the WhatsApp Business (Cloud) API - through a Business Solution Provider like TAM Inbox, or directly - the agent isn't just a Meta Business Manager checkbox. It has a real, documented API surface, and understanding its shape matters if you want to know what actually happens to your numbers and your existing inbox when you turn it on.

Meta's own developer documentation lays this out for two integration patterns. A business managing its own WhatsApp Business Account (WABA) directly can configure the agent using its own system-user token. A platform operating WhatsApp on behalf of multiple client businesses - what Meta calls a Business Solution Provider or Tech Provider, which is what a platform like TAM Inbox is - can configure and manage the agent on a client's WABA using a business-integration system-user token, the same delegated-access model already used for template and phone number management.

Configuration happens at the phone number level within a WABA, not account-wide. The flow, per Meta's documentation, moves through distinct stages: checking a number's eligibility, onboarding it (assigning the WABA to the agent), configuring it (skills, business info, FAQs, connected websites, uploaded files, and third-party connectors), and then operating it day to day (thread control, agent events, testing and evaluation endpoints). Each stage is its own set of API calls, not a single on/off switch.

The part that matters most for anyone already running a team inbox: once the agent is live on a number, your own platform doesn't lose visibility into the conversation. Meta places your integration in what it calls standby - you keep receiving copies of the messages the agent is handling - and there's a dedicated Thread Control endpoint that lets you explicitly take a conversation back (Meta calls this a "pass" action) the moment a human needs to step in. That's the mechanism that makes "AI handles the easy stuff, a person handles the rest" actually work operationally rather than just as a slide in a pitch deck, since nothing about it requires the customer to start a new conversation or the agent hand-off to be manual on Meta's side.

On cost: Meta has published API-connected pricing at $2.00 per 1 million tokens, effective August 1, 2026 - a genuinely different cost model from the per-conversation pricing that governs regular template and session messages, so it's worth budgeting for separately rather than assuming it folds into your existing conversation costs. Worth confirming current rates directly with Meta before committing volume, since token-based AI pricing is a newer, faster-moving line item than the conversation pricing this blog has covered before.

If you're evaluating whether to build against this API surface yourself versus waiting for your platform to support it: the honest answer is that eligibility, onboarding, and eligibility-gating by country and vertical mean this isn't yet a flip-a-switch feature for every business - test the Eligibility endpoint for your specific number before assuming it's available, and treat the Skills/FAQs/Connectors configuration step as the real work, since an agent is only as good as what it's been given to work from.